文献类型：期刊文献

英文题名：A zeroth-order variance-reduced method for decentralized stochastic non-convex optimization

作者：Chen, Hongxu[1];Chen, Jinchi[2];Wei, Ke[1]

机构：[1]Fudan Univ, Sch Data Sci, Shanghai, Peoples R China;[2]East China Univ Sci & Technol, Sch Math, 130 Meilong Rd, Shanghai 200237, Peoples R China

年份：2025

外文期刊名：OPTIMIZATION

收录：;WOS:【SCI-EXPANDED(收录号:WOS:001634053100001)】；

语种：英文

外文关键词：Decentralized non-convex optimization; zeroth-order method; variance reduction; stochastic optimization; linear speedup

摘要：In this paper, we consider a distributed stochastic non-convex optimization problem, which is about minimizing a sum of n local cost functions over a network with only zeroth-order information. A novel single-loop Decentralized Zeroth-Order Variance Reduction algorithm, called DZOVR, is proposed, which achieves $ \mathcal {O}(dn<^>{-1}\epsilon <^>{-3}) $ O(dn-1 & varepsilon;-3) sample complexity at each node to reach an & varepsilon;-accurate stationary point and also exhibits network-independent and linear speedup properties. To the best of our knowledge, this is the first stochastic decentralized zeroth-order algorithm that achieves this sample complexity in smooth setting. Numerical experiments demonstrate that DZOVR outperforms the other state-of-the-art algorithms and has network-independent and linear speedup properties.